For a long time now, people have been asking whether and how artificial intelligence (AI) can be created, but the topic has never been so hotly discussed as it is today. Whatever the sector, the general feeling seems to be that AI is on the brink of a breakthrough and set to make our lives a whole lot easier and better.
Our quest for artificial intelligence is based on our wish to enhance mankind’s cognitive skills with even better intelligence, to thereby optimise our own opportunities and to achieve better and faster work results. In doing so, we are following mankind’s own will to cross new borders and see what lies beyond them. According to historian and author Yuval Noah Harari, man is instinctively driven to achieve eternal life and therefore hopes to become godlike. Artificial intelligence brings us one small step closer to this goal.
Or that’s the great hope. A hope that is of course extolled by developers and supporters. On the other hand, you have the critics pointing out that the status quo isn’t that advanced yet anyway, or those who fear that AI could cause apocalyptic dangers. As a visionary of our time, Elon Musk represents the critics and admonishers. Artificial intelligence as the potential trigger of a third world war is the most drastic prospect for the future as described by Musk. The truth often lies somewhere in between, and it is certainly true that politicians and social researchers should now also be dealing with the issue in more depth and developing regulations for the future, not waiting until companies take advantage of the new technologies without legal restrictions. This year the United Arab Emirates was the first country in the world to set up a ministry for artificial intelligence: they wanted to be the country that is best prepared for this new global wave. Russian President Vladimir Putin also seems to be prioritising the subject very highly. Ideally, such pioneers should not be the only ones, lots of international political supporters would be better.
Regardless of however we look at it ourselves: the development cannot be stopped, possibly only temporarily regulated. But where are we at today? How far has artificial intelligence come?
To answer this question, we first have to define more precisely what we understand by AI.
What is artificial intelligence?
No doubt everyone has their own idea that instantly springs to mind. But an objective definition can of course be found on Wikipedia, which says that artificial intelligence “is intelligence displayed by machines, in contrast with the natural intelligence (NI) displayed by humans and other animals. (…) Colloquially, the term “artificial intelligence” is applied when a machine mimics “cognitive” functions that humans associate with other human minds, such as “learning” and “problem solving”. The field was founded on the claim that human intelligence “can be so precisely described that a machine can be made to simulate it”.” So AI leaves a certain amount of room for interpretation.
All too often, however, we are dealing less with intelligent systems in the sense of human intelligence and much more with the rapidly developing possibilities of machine learning. Or to put it simply, machine learning is the artificial generation of knowledge from experience. Computers are built in such a way that they can use self-learning algorithms to identify patterns or correlations in data, draw conclusions and make predictions on their own so the findings gained from the data can be generalised and used for problem solving.
In order for computers to learn independently and find solutions, IT systems first have to be provided with relevant data and algorithms. To recognise patterns, rules also have to be created for the analysis of data.
Machine learning is currently regarded as one of the central and most successful sub-disciplines of artificial intelligence.
Although machine-learning methods are still based on research from the 1980s, they are becoming more widespread with technical progress. With new database technologies, significantly more powerful processors and improved algorithms, machine learning is also becoming increasingly popular.
The biggest development stages of machine learning are man-versus-machine competitions that demand the highest cognitive performance of IT systems. At the beginning of 2016, Google’s AlphaGo machine-learning system easily outclassed the reigning World Champion of the Asian strategy game Go, winning four out of five rounds.
The AlphaGo algorithm also helped Google to reduce the energy consumption of its data centre by 15%. This example shows the very high economic relevance of machine learning or artificial intelligence.
Of course these processes are impressive. However, we have only made a small step forward in producing intelligence that is comparable to that of human beings, and we are only scratching the surface of true artificial intelligence.
But we can confirm that with artificial intelligence we are already able to make human performance much more efficient, productive and even more creative. So artificial intelligence, i.e. today’s machine-learning methods, complement human skills in an incredibly valuable way.
The possibilities of use in our everyday life seem to be endless, for example in the analysis of medical studies, in logistical evaluations to optimise delivery processes or when analysing measurements for failure predictions in engineering.
What is the status quo from a digital agency’s perspective?
The topic of artificial intelligence is also omnipresent in our everyday agency life. And no longer just in the classic organisation and editing of images, but also in areas including marketing automation, customer relationship management, chatbots and image and object recognition.
Machine-learning systems are already unleashing their potential today, especially in the field of marketing automation. Here, systems are required that analyse digital marketing processes and perform repetitive marketing tasks automatically. This affects, for example, the rule-controlled and personalised delivery of emails, messages, videos, social media posts, website contents, shop offers or advertisements. With the support of machine-learning procedures, we continuously analyse how the distribution of communication and contents can be optimised, depending on the marketing objective and consumer reaction. This is no longer possible with an Excel spreadsheet, but requires powerful systems that speed up and improve classic marketing.
It quickly becomes clear that the art lies in approaching a technological and analytical infrastructure as holistically as possible. This will allow artificial intelligence to learn from all end customer contact points, therefore supporting the digital marketing team in fully understanding customer journeys and being able to fully evaluate and optimise touchpoints with individual end customers or end customer segments. Experience shows that a certain system or provider would not suffice for this, despite what various providers are proclaiming with their marketing clouds. Depending on its specific goals and requirements, a marketing organisation has to create its own modular, but networked technological and analytical infrastructure, along with the corresponding processes. This would then enable huge successes in digital communication – reflected in a considerably higher level of productivity. This means that the marketing organisation considerably increases its resources and, in combination with a higher level of adaptability and scalability, also its competitiveness. Such investments result, above all, in noticeably better conversion rates from a customer contact point to an economically valuable transaction. In our experience, increases of up to 35% are not uncommon.
Customer relationship management
We see enormous potential in the CRM sector in particular. Chatbots, for example, have developed at record speed from a trend topic into a nearly established marketing instrument. After the first prototypes, we are already regularly using them for our clients as a dialogue instrument. But all too quickly, chatbots are being put on a level with intelligent systems that need to have smart answers to all questions. First of all, this is a fallacy. Chatbots first have to be fed with lots of functions, information and reaction options. No bot will be able to communicate perfectly to start with. Unfortunately, the risk that both the sender and recipient of messages will be disappointed with the system’s performance is too high.
This can be prevented by setting up the systems as expected from the beginning, not promising too much and, above all, using closed questions with predefined answer options. While the bots are in use, it is also essential to continuously analyse consumer dialogues and successively optimise the system, making it increasingly smarter and accurate in its answers.
By integrating additional external data sources such as the weather, news, product updates etc., the usefulness can also be optimised step by step. As well as the chance to train chatbots and network them with different data sources, in the future it will also be easier to develop semantic systems that are in a position to understand a growing number of idiomatic expressions and therefore provide more accurate answers – another step towards intelligent systems.
The MINI John Cooper Works chatbot
Nothing is easy to start with. We have not yet become accustomed to the fact that we are using more and more public beta solutions in these dynamic times of digitalisation to optimise systems faster and in a more goal-oriented way. With the release of public beta versions, Apple is a good example of this. These pre-release versions used to be reserved for developers only. But there are lots of interested end consumers today who will be happy to support improvements to the system using feedback functions.
In addition to the singular use of individual chatbots, these dialogue robots will hugely advance the optimisation of existing CRM systems. The more intensively and regularly chatbots are used, the more essential it will be to record all customer touchpoints in one system and to synchronise each dialogue status with the consumer. In the future, a bot will have to know that there has been a prior telephone call and e-mail correspondence with the customer service team, and also which products the buyer has purchased from the brand in order to be able to process customer complaints faster and of course also to boost upselling and cross-selling. Always with the goal of only giving the consumer the information or offer that actually interests them at the right moment, and protects them from the danger of information overload. The key term here is ‘moment marketing’. The assumption is that many customers would feel less inhibited asking a machine questions than they would a consultant. After all, you can’t embarrass yourself in front of a machine. And in the finance sector especially there are new and interesting findings: according to a recent global study by Accenture with 33,000 respondents in 18 countries, rather than scepticism, there is a great deal of trust in an objectively operating investment consulting machine – a so-called ‘robo advisor’. Accordingly, 78% would like to make use of this service.
Image recognition and machine learning
We have been concerned with image and object recognition systems for a long time now, especially in the mobile environment. Even before the market launch of the iPhone, we used image recognition systems to network out-of-home billboard campaigns that at the time were used for robotics purposes. Billboard images were photographed by passers-by with the feature phone and sent to a server via multimedia message (MMS). The system recognised the image and answered with image-specific feedback. That was in the year 2005 and it was astoundingly precise. Overall an exciting function, but with a low usage – as is so often the case when you are breaking new ground. The usability was still a long way away from real user-friendliness. But this was also during a time when the average consumer couldn’t imagine that we would soon be surfing the internet when we were out and about. And why should they have?
Today, for example, Apple’s new ML interface allows us to find comparable systems like Core ML:
here we can see very clearly how well developed the field of machine learning is. In milliseconds, the database tries to recognise identified objects that appear in front of the camera. This happens very quickly, but the accuracy is still not perfect. However, this is set to change. All the big tech companies are working on corresponding solutions and it won’t be long until the accuracy is improved either. Eventually it will be possible to deliver location and context-related information via camera functions, regardless of the hardware being used. So with machine learning skills, augmented reality will herald a new information era.
Voice assistants and natural language processing
The Amazon Echo recently started the ball rolling and paved the way for a huge offer of voice assistants. Amazon is offering an ever-growing product variety, Google is entering the race with Google Home, Apple is launching the Homepod and Samsung is releasing Bixby as a voice assistant in its systems. And all other providers like Sony, Panasonic and Onkyo are providing new hardware and picking up on the available voice systems of the big players.
With this trend, we will gradually learn how to talk to machines. Users often start with high expectations and then soon notice that the systems can’t understand everything after all and are unable to answer every question. So here we are seeing the same phenomenon as with the chatbots. But with increasing use this will also change. The systems, which ultimately have a similar structure to chatbots, are also fed with increasing amounts of data, gradually trained and semantically enhanced.
Language will definitely be one of the exciting interfaces of the future that people will use in conjunction with technology. Over time, the assistants will chat more naturally with the users and seem more intelligent, especially as they are responding more and more to moods on the basis of voice analysis, meaning that they have what it takes to become empathetic family members!
Brands are well advised to already start gaining their first experience, to test possible areas of use and to firmly integrate this instrument into their digital innovation landscape. After all, what could be better for a brand than to become a likeable, friendly member of the target group’s family as well as a brand ambassador, advisor and sales assistant all rolled into one?
How should we deal with artificial intelligence?
No worries, despite all these developments, we humans will still remain a valuable resource. In our tasks, however, we can and should focus more on the emotional aspects, creativity and strategy. After all, that is something machines won’t be able to do so quickly. We are still years away from them having a human-like intelligence.
But: artificial intelligence or – per status quo – rather machine learning, already has enormous potential today. For all industries and particularly for marketing and communication. It is important to gain experience together quickly and therefore actively shape the development process.
How can you create the conditions to experiment with artificial intelligence yourself?
Providers like Microsoft, IBM, Google and Amazon are offering platforms and services for the development of intelligent applications. This allows developers to build intelligent applications without special machine-learning knowledge, which learn from a freely selectable database. At IBM this platform is called Watson, at Microsoft Azure ML Studio, at Amazon it is known as Amazon Machine Learning and at Google it goes by the name of Tensorflow.
The most common and popular are free, high-quality open source solutions, which make artificial intelligence accessible to a wide audience of data scientists and developers. When it comes to the programming languages, in this context Python is the favourite. R and Java are also popular. The currently most popular database or data processing systems include MongoDB, Neo4j and Apache Spark. The dynamic development of such offers is enormous. It is interesting that providers like IBM and Google also use open-source components that they integrate into their portfolio.
In our agency group we are also using a system of open-source components. From these, we have set up our own platform called CORE, for the flexible development of machine learning solutions. Specifically recruited data science specialists work together with communications and digital specialists and our clients to develop solutions for different problems together on the CORE platform. This is how we manage, step-by-step, to make marketing communication programmes much more efficient for our clients.
With the BrandCube for example, we were able to develop an instrument that makes communication investments in individual marketing objectives predictable and plannable in individual target segments. In just a few minutes, algorithms use extensive empirical data to calculate how a marketing communication budget should be invested most efficiently.
This enables our agency specialists to quickly create investment scenarios and coordinate them with our clients when required. Such machine-learning-supported processes have already led to insights into how marketing communication and sales investments can be combined to optimise each other to such an extent that the economic result of the marketing communication increased by more than 14%.
Such significant benefits are possible with the participation of artificial intelligence or machine learning methods in close and transparent cooperation with all specialists involved in the process that needs optimising. They would communicate on a shared platform and iteratively approach a common vision or solution to a problem.
So let’s join forces to shape the future and this transformation together.
Illustration: Emiliano Ponzi